Thai Journal of Operations Research : TJOR

Developing a predictive model for student learning success by applying ensemble learning

2024-07-19T08:55:39+07:00

This research applies ensemble learning, which combines multiple models to analyze student learning. The objectives are: 1) to develop a model for predicting the academic success of 397 M.4 students in the second semester of the academic year 2023 at Srinakharinwirot University Prasarnmit Demonstration School (secondary), who are enrolled in the mathematics course 2(1) M31208. The prediction uses student learning behaviors observable by teacher during learning activities, comprising five aspects of learning behavior: learning ability, responsibility, class participation, diligence, and determination; and 2) to analyze the factors that significantly impact students' academic success. The study compares the performance of five classification models: Elastic–net regularized generalized linear model (GLM), Random Forest (RF), Gradient Boosting with XGBoost (XGBoost), Artificial Neural Network (ANN), and ensemble learning with Stacking (stack), evaluating them using F1 score, recall, and precision. The results indicate that the RF and XGBoost models have the highest performance (F1 score = .904, recall = .891, and precision = .920). Additionally, learning ability, represented by average quiz scores, is identified as the most significant factor for academic success (importance = 153.818), while responsibility, represented by the proportion of bringing books to class, is the least significant factor (importance = 1.660).

Development of a genetic algorithm for partner selection in business networks with or without intermediaries

2025-05-06T08:36:52+07:00

This research investigates the problem of business partner selection within a business network, where some organizations can invest in partnerships directly, while others must invest through intermediaries. The remaining organizations have the flexibility to invest either directly or via intermediaries to maximize overall benefits under a limited budget. This problem is highly complex and computationally challenging. The study begins by proving that the problem is NP-hard, highlighting its computational complexity and justifying the application of genetic algorithms as a suitable approach. Subsequently, an exact optimization algorithm is developed alongside the implementation of a genetic algorithm. The results indicate that the exact optimization algorithm is effective for small business networks with no more than 14 organizations. A performance comparison between the two methods reveals that the genetic algorithm produces solutions consistent with those obtained from the exact optimization algorithm on the same test instances while requiring significantly less computational time. Furthermore, this research conducts an experimental comparison between the genetic algorithm, particle swarm optimization (PSO), and the exact optimal solutions. The results indicate that the genetic algorithm consistently produces accurate solutions across all tested cases. However, the PSO approach exhibits minor inaccuracies, with errors not exceeding 15%. A statistical t-test (t = 1.497, p = 0.145 > 0.05) suggests no statistically significant difference between the two methods. Nonetheless, in terms of computational efficiency, the genetic algorithm significantly outperforms the PSO approach. To further demonstrate the suitability of the genetic algorithm for large-scale business networks, this study evaluates its performance on networks with 60 to 100 nodes. The results confirm that the genetic algorithm consistently provides accurate solutions within a reasonable computation time. Consequently, the genetic algorithm is deemed an appropriate approach for solving this problem in large and complex business networks.

Inventory cost reduction for spare parts case study for dry pet food production company

2025-04-25T14:07:14+07:00

Inventory management played a crucial role in reducing costs and enhancing production efficiency. This study examined inventory management using classification methods such as FSN Analysis and ABZ Classification. The research employed the OUL-T Model and QR Model to determine optimal inventory levels, aiming to improve spare parts management and reduce associated costs. The study was based on a case study of a company operating in Thailand, focusing on analyzing five years of historical data and applying Monte Carlo simulation to forecast future demand. The findings revealed that classification and demand forecasting significantly enhanced inventory management accuracy and reduced the cost of excess inventory by 543,139 THB. Furthermore, the results of this research could be adapted for other industries to improve inventory management efficiency.

Understanding what do Thai people think about different kinds of COVID-19 vaccines using text mining

2025-04-23T10:15:23+07:00

The research aimed to apply text mining techniques to analyze the opinion of Thai people about the Covid-19 vaccine. Thai Twitter data used in this study was searched through keywords related to Covid-19 vaccine brand names: AstraZeneca, Sinovac, Sinopharm, Pfizer, and Moderna from March to October 2021. There were 451,209 tweets. All the tweets collected were separated on Covid-19 vaccine brand names and then classified topics using with Latent Dirichlet Allocation (LDA) technique. Next, sentiment analysis of the tweets on different topics was done using the Multinomial Logistic Regression Model. As a result, tweets related to the AstraZeneca vaccine can be classified into 8 topics. Most of them illustrated the Mix and match of different Covid-19 vaccines on Thailand’s vaccination policy, including the side effects. Sinovac vaccine can be classified into 9 topics. Most of the tweets stated that the celebrity getting vaccinated. Sinopharm vaccine can be classified into 10 topics. Most of the tweets talked about the problems of vaccine booking systems. Pfizer vaccine can be classified into 10 topics. Most of the tweets discussed the delay or shortage of place orders for vaccines, whereas Moderna vaccine can be classified into 4 topics. Most of them focused on the problems with the vaccine booking system. In addition, it revealed that most Thai people’s opinions on social media are neutral polarity. Negative sentiment tweets constituted a larger share as compared to positive sentiment tweets about vaccines. Most of the topics in the tweets with negative sentiment involve insufficient vaccines, Covid-19 vaccine procurement management by the government Issues, the online systems of the alternative vaccines defined in Thailand, including side effects after vaccination, and even reviews of vaccinations of Thai celebrities or influencer people.

Developing selection criteria for mechanical, electrical and plumbing custom-made product suppliers in construction projects using the fuzzy analytical network process

2025-04-24T14:02:15+07:00

The research applies a multi-criteria decision-making method to select suppliers in the construction industry, specifically for custom-made material and equipment suppliers. The initial criteria were gathered from a review of academic articles, and the Delphi technique was used to collect expert opinions confidentially in order to determine the evaluation criteria. The weights of each criterion were then analyzed using the Fuzzy Analytical Network Process (FANP), as fuzzy set theory effectively handles uncertainty. Furthermore, the Analytical Network Process enables a comprehensive evaluation by considering the relationships between sub- criteria at each level or within the same level. The research findings revealed that the evaluation criteria for the case study's selection process include 6 main criteria and 17 sub-criteria, covering key success factors for construction projects in terms of price, quality, transportation, after-sales service, required operating condition and supplier’s organizational structure, respectively.

A comparative analysis of gold price forecasting efficiency using model averaging over regression trees and random forest

2025-04-24T09:14:37+07:00

This research aims to compare the forecasting performance of gold prices using two tree-averaging methods: Multi-Model Tree Averaging and K-Fold Tree Averaging. Both methods employ weighting schemes, namely Equal Weight and AIC Weight. The study also compares these methods with Random Forest to evaluate forecasting accuracy and stability. The data used are monthly secondary data on economic variables, such as inflation rates and interest rates in Thailand, the United States, China, and India, spanning from January 2014 to July 2024. The results show that Random Forest achieves the most suitable performance in terms of R², RMSE, and MAPE in both training and testing datasets, demonstrating superior accuracy and stability. Multi-Model Tree Averaging and K-Fold Tree Averaging rank slightly lower in performance, with AIC Weight consistently outperforming Equal Weight in all scenarios. This study highlights the potential of diverse forecasting methods and provides recommendations for selecting appropriate techniques for predicting gold prices and other highly volatile assets.

Bead production scheduling with setup time: a case study of a truck tire factory

2025-06-10T18:01:42+07:00

This research studies and compares mathematical models for production scheduling of bead production in truck tire factory. The objective is to reduce excessive inventory and improving labor efficiency, both of which are critical factors influencing overall production cost. The bead production process involves three stages, with 23 workers. The process produces six types of bead with a total of 13 machines. Three operational workers are responsible for managing inventory and production scheduling separately in each process. Scheduling is based on customer demand and current stock levels which aim to prevent shortages, but it often leads to overproduction and inefficient use of labor. To address these challenges, three mathematical models were developed: 1) minimizing machine setup frequency, 2) minimizing machine operation days, and 3) minimizing the number of storage carts. A comparison between model outputs and actual production data shows that all three models can reduce inventory levels. However, labor costs remain a major concern. Among the models, the second demonstrates the most promising results by reducing labor usage by 14.13 percent and machine setups by 27.23 percent. Thus, the second model is selected as the most suitable for production planning in the case study factory.